绿色算力定义及关键技术研究

绿色算力是实现算力基础设施发展与生态环境保护有机统一的综合能力,是衡量算力基础设施发展绿色化程度的一个综合指标。绿色算力是把自然资源、环境资源作为算力发展考虑的关键,旨在推动算力设施、算力设备、算力平台和算力应用的“四位一体”绿色发展。绿色算力的目的是打通算力供需通路,完成算力产业用能转型、生产高效、应用和需求场景相匹配。通过研究绿色电力、绿色计算的概念,结合绿色发展战略规划的相关内涵,总结得出绿色算力的定义。绿色算力技术是保障绿色算力发展的技术体系,我国绿色算力技术体系正不断发展与完善。梳理绿色算力关键技术,为后续政策研究、标准制定、推广绿色算力消费和引导算力产业整体绿色高质量发展打下研究基础。     

算力网络四面三级算力度量技术体系

算力网络要实现广泛的算力互联，就离不开统一的算力度量和算力建模机制。算力度量技术主要解决算力描述的标准化，提供方便的跨厂商互通能力、算力资源协同管控能力。对算力网络中算力度量需求进行分析，并提出四面三级算力度量技术体系，同时还探索了在算力路由决策中，网络按需进行算力信息通告等关键技术，为后续的算力度量研究提供参考。     

面向算网一体化演进的算力网络技术

随着一体化算力网络国家枢纽节点的建设和“东数西算”工程的加快实施,计算和网络的融合走向深水区。面向新兴业务对于网络和计算融合发展的需求,如何协调分布式、多样化算力资源,业务应用和网络资源满足业务需求成为亟待解决的问题。为了解决计算和网络相互感知、协同、调度问题,从算力网络阶段化发展、算力感知技术架构、算力度量与标识、算力路由等多个技术展开探索和研究,并提出算力网络多种部署模式,为算力网络后续技术研究和产业发展提供参考。     

东数西算场景下的算力网关研发及应用

提出了一种面向算力网络场景的新型网络设备——算力网关。认为算力网关是实现算力网络一体化调度的基础，通过感知业务应用需求，结合当前的算力状况和网络状况，生成路由信息并发布到网络，将计算任务报文路由到合适的计算节点，以实现算力资源的最优调度。现网实践验证了算力网关技术方案的有效性。     

算力调度关键要素及路径分析

我国算力规模已在全球排名第二,近五年平均增速超过30%,高速增长的背后需要考虑如何满足千行百业的需求。本文从算力及算力应用场景分类入手,针对不同场景分析总结了影响算力调度的关键要素及策略,进一步对不同应用可采取的调度路径进行了分析研判,旨在提供一个有效使用算力的思路。     

从“算力中心”到“算力网”——从算力角度谈算网一体的机遇与挑战

文章以算力的主视角切入,探讨算网一体概念对算力基础设施建设的重要影响,重点分析由“算力”向“算力网”发展过程中所面临的关键技术挑战,并面向该系列挑战提出一套算力网基础功能架构,开展应用案例分析,为“算网融合”的概念演进与技术发展提供关键参考。     

综合算力发展现状与趋势分析

随着数字经济的快速发展,综合算力作为支撑各种应用和服务的新型生产力,其重要性日益凸显。首先,对综合算力的概念、重要性以及发展现状进行了分析和研究;其次,从算力形式、运力性能、存储技术等维度对综合算力的发展趋势进行了展望;最后,提出了推动综合算力高质量发展的相应建议。     

算力网络发展中的若干关键技术问题分析

介绍了目前国内外算力网络技术的发展现状,对算力网络关键技术进行了阐述,从算力度量与算力建模、基于算力信息的路由决策、云-边-端算力协同、基于服务的云网融合及算力网络信息安全5个方面,对算力网络技术发展及实践应用中遇到的问题进行了深入分析,并提出了初步的解决方案。     

高速公路复合算力需求与算力路由供给

算网融合成为信息通信行业发展目标,在此背景下,文章分析未来信息通信行业发展趋势和高速公路复合算力需求,探讨高速公路算力路由趋势。研究结果表明算力路由是算网融合的基础条件,高速公路是建设算力路由的有效载体,不仅可为高速公路的本地算力需求形成便捷的算力路由供给,而且可将积累的闲散算力错峰供给城市其他行业创造经济效益。算力路由建设将有效提升高速公路信息化管理水平、降低运营维护成本和创造新增效益,为未来算网融合提供新型思路。     

算力网络应用平台研究与设计

针对算力网络（CPN）因算力多样泛在、多要素融合、需求差异化的特点导致的用户使用困难、应用管理难等问题,从纳管与接入、决策与调度、应用与开放三个层面研究算力网络应用的管理以及异构算力融合和跨域编排关键技术,并设计一个算力网络应用平台。通过提供一种即开即用、按需付费的零感知算网应用服务,使用户关注于应用服务需求,无需关心算力资源需求和算力网络复杂环境,从而实现降低用户的算力网络使用门槛,提升算网应用的使用和管理效率的目标。     

SDR receiver computational model for application in power control

Software-defined radio (SDR) permits dynamic switches of the employed radio access technology (RAT), over-the-air (OTA) software updates, software and hardware reuse. This extended flexibility comes at the price of a higher computing complexity and, in particular, the energy consumption at the receiver. The analysis of the computational profile of signal processing algorithms is of great importance in SDR for understanding the implication on the energy consumption. Several signal processing algorithms show a different profile as a function of the signal quality perceived at the receiver antenna. Therefore, power control policies have an implication on the computational performance of SDR receivers. Understanding the behaviour of these algorithms allows trading transmitted power against receiver energy consumption. This paper presents a model for characterizing the computational profile of Turbo and LDPC decoders and demonstrates is applicability in existing power control strategies.     

Early computational electromagnetics

The author discusses the early history of computational electromagnetics from the late 1940s. In particular Illiac and Ordvac are discussed     

可信计算概论

随着基于互联网的应用系统不断增多,人们所面临的安全问题也与日俱增。如何构建新一代适应信息发展需求的可信计算环境已经成为信息科学技术领域最重要的课题之一。     

The computational eye

Specialized cells in the eye work in parallel for unequaled image processing and computation. Millionfold swings in light intensity from the outside world, transformed to electrical signals, are processed in space and time for contrast and edge enhancement, as well as to detect motion     

An efficient computational method for characterizing the effects of random surface errors on the average power pattern of reflectors

Based on the works of Ruze and Vu, a novel mathematical model has been developed to determine efficiently the average power pattern degradations caused by random surface errors. In this model, both nonuniform root mean square (rms) surface errors and nonuniform illumination functions are employed. In addition, the model incorporates the dependence onF/Din the construction of the solution. The mathematical foundation of the model rests on the assumption that in each prescribed annular region of the antenna, the geometrical rms surface value is known. It is shown that closed-form expressions can then be derived, which result in a very efficient computational method for the average power pattern. Detailed parametric studies are performed with these expressions to determine the effects of different random errors and illumination tapers on parameters such as gain loss and sidelobe levels. The results clearly demonstrate that as sidelobe levels decrease, their dependence on the surface rms/lambdabecomes much stronger and, for a specified tolerance level, a considerably smaller rms/lambdais required to maintain the low sidelobes within the required bounds.     

电磁计算方法研究进展综述

文章概要介绍了电磁计算方法的研究进展.首先对电磁计算方法的发展进行了概述.其次,对近些年发展出来的若干代表性电磁计算技术,包括快速直接法、非共形区域分解法、高性能并行技术等的发展进行了阐述.再次,对典型电磁计算问题,包括地海复合目标、大规模有限周期结构、电磁逆问题等电磁计算技术的发展进行了简要阐述.最后,对电磁计算方法的发展进行了总结和展望.     

Dynamic algorithms in computational geometry

Dynamic algorithms and data structures in the area of computational geometry are surveyed. The work has a twofold purpose: it introduces the area to the nonspecialist and reviews the state of the art for the specialist. Fundamental data structures, such as balanced search trees and general techniques for dynamization, are reviewed. Range searching, intersections, point location, convex hull, and proximity are discussed. Problems that do not fall into these categories are also discussed. Open problems are given     

Parallel techniques for computational geometry

A survey of techniques for solving geometric problems in parallel is given, both for shared memory parallel machines and for networks of processors. Parallel models are reviewed, and basic subproblems that tend to arise in the solution of geometric problems on any parallel model, are discussed. PRAM techniques, techniques for mesh-connected arrays of processors, and the hybrid RAM/ARRAY model and its connection to I/O complexity are considered. Open problems are also discussed, as well as directions for future research     

Approximate correctness-checking of computational results

The process of checking the correctness of time-critical computations involves a fundamental and inherently conflicting tradeoff: the precision of the checking vs the time required to perform the checking process. This tradeoff is particularly relevant when the computations are intensive and can require appreciable time for completion. This issue is addressed by an innovative technique based on an extension of the `use of certificates' which permits the precision of the correctness checking to be explicitly specified. This concept is developed relative to the broad class of problems which are amenable to the powerful and widely used B&B (branch-and-bound) strategy. It is extremely relevant to the motivation and importance of this paper to remember that `checking the correctness of results for computationally intensive problems' can be just as demanding as `generating the results themselves'. Accordingly, for time-critical applications in which a run-time result verification must be accomplished before a generated result can be released, a method for highly efficient approximate correctness checking of results in which it is efficiently determined that an obtained result is within some specified percentage of an optimal solution can be of enormous importance. This paper presents an approximate correctness checking methodology in which the precision of the result check can be explicitly specified and guaranteed even when an exact correct result is unknown, This controllable tradeoff, inherent in this method, can be the basis for new classes of adaptive result-checking strategies in which the precision of the check of a result can be varied, depending on the impact of the result within time-critical applications     

Multi-grid interface in computational electromagnetics

An elegant and accurate technique of interfacing segments of a uniform computational mesh with different spatial resolution is described. The technique is illustrated in the time-domain for a TLM mesh. Results show that a seamless interface is achieved between different mesh areas, thus allowing multi-scale problems to be accurately mapped onto the mesh.